AI researchers are great with numbers — just not numbers with dollar signs in front of them

By Alan Jacobson, Systems Architect & Analyst

The AI industry is optimizing the wrong scoreboard, and the consequences are starting to show up in public.

The industry keeps celebrating the wrong wins

For the past two years, AI “progress” has been defined almost entirely by research metrics:

benchmark scores
capability demos
scale milestones
technical firsts

All of that matters — early.

But none of it answers the question that now dominates reality:

What does this cost to run, and how do those costs behave at scale?

This isn’t a technical failure. It’s a leadership mismatch.

Look at who is running the most influential AI labs:

Anthropic
OpenAI

Both are led by researchers.

That’s not a character flaw. It’s a structural one.

Researchers are trained to optimize:

accuracy
capability
theoretical limits
performance curves

They are not trained to optimize:

unit economics
margin durability
cost compounding
long-term P&L behavior

So the system keeps doing what it knows how to do — even as the problem changes.

The business model was always supposed to “come later”

That assumption is now breaking.

You can see the strain in plain sight:

Subscriptions don’t map cleanly to operating cost
Advertising is floated as a solution, despite poor fit
“Value sharing” is discussed without clear metering or enforcement
Revenue narratives keep changing, but costs remain stubborn

These are not strategies.
They are symptoms.

The missing discipline: dollars per interaction

Most AI decisions are still evaluated without a dollar sign attached.

What gets measured:

performance
accuracy
user delight

What often doesn’t:

marginal cost
repeatability
cost compounding across massive volume

As a result:

impressive demos coexist with fragile economics
scale increases usage faster than profitability
costs rise invisibly until they don’t

That’s why, despite unprecedented attention and adoption, there are still remarkably few beans to count.

This is what happens when research leads past the research phase

Every industry goes through this transition:

Early phase: breakthroughs win
Middle phase: adoption explodes
Mature phase: economics decide everything

AI is crossing that boundary right now.

When research culture remains in charge too long:

optimization targets lag reality
economic discipline arrives late
leadership keeps solving yesterday’s problem

That gap is now becoming expensive.

The next winners won’t look like the last winners

The next phase of AI will favor companies that can:

explain their economics clearly
show improving margins with scale
control costs without degrading utility
speak credibly to CFOs and investors

Not just other researchers.

That doesn’t mean research stops mattering.
It means research stops being enough.

The bottom line

The AI industry is full of brilliant people doing brilliant work.

But brilliance alone doesn’t pay the bills.

Until leadership starts optimizing for numbers with dollar signs in front of them, the industry will keep producing stunning technical achievements — and an uncomfortable lack of financial ones.

And that’s a gap the market will eventually close, whether the industry is ready or not.